diff --git a/pkgs/development/compilers/adaptivecpp/default.nix b/pkgs/development/compilers/adaptivecpp/default.nix
index 6cb32e3e92d2..dc5f69e71e9c 100644
--- a/pkgs/development/compilers/adaptivecpp/default.nix
+++ b/pkgs/development/compilers/adaptivecpp/default.nix
@@ -2,7 +2,6 @@
   lib,
   fetchFromGitHub,
   llvmPackages_17,
-  lld_17,
   python3,
   cmake,
   boost,
@@ -18,23 +17,40 @@
   cudaSupport ? config.cudaSupport,
   autoAddDriverRunpath,
   callPackage,
+  symlinkJoin,
   nix-update-script,
 }:
 let
   inherit (llvmPackages) stdenv;
   rocmPackages = rocmPackages_6;
   llvmPackages = llvmPackages_17;
-  lld = lld_17;
 in
 stdenv.mkDerivation (finalAttrs: {
   pname = "adaptivecpp";
-  version = "24.06.0";
+  version = "24.10.0";
 
   src = fetchFromGitHub {
     owner = "AdaptiveCpp";
     repo = "AdaptiveCpp";
     rev = "v${finalAttrs.version}";
-    sha256 = "sha256-TPa2DT66bGQ9VfSXaFUDuE5ng5x5fiLC2bqQ+ZVo9LQ=";
+    sha256 = "sha256-ZwHDiwv1ybC+2UhiOe2f7fnfqcul+CD9Uta8PT9ICr4=";
+  };
+
+  # we may be able to get away with just wrapping hipcc and nothing more
+  # this is mainly so that if acpp tries doing <PATH_TO_HIPCC>/../amdgcn/bitcode
+  rocmMerged = symlinkJoin {
+    name = "rocm-merged";
+    paths = with rocmPackages; [
+      clr
+      rocm-core
+      rocm-device-libs
+      rocm-runtime
+    ];
+    buildInputs = [ makeWrapper ];
+    postBuild = ''
+      wrapProgram $out/bin/hipcc \
+        --add-flags "--rocm-device-lib-path=$out/amdgcn/bitcode"
+    '';
   };
 
   nativeBuildInputs =
@@ -57,23 +73,24 @@ stdenv.mkDerivation (finalAttrs: {
       llvmPackages.libclang.dev
       llvmPackages.llvm
     ]
-    ++ lib.optionals rocmSupport [
-      rocmPackages.clr
-      rocmPackages.rocm-runtime
-    ]
     ++ lib.optionals cudaSupport [
       cudaPackages.cuda_cudart
       (lib.getOutput "stubs" cudaPackages.cuda_cudart)
     ];
 
   # adaptivecpp makes use of clangs internal headers. Its cmake does not successfully discover them automatically on nixos, so we supply the path manually
-  cmakeFlags = [
-    "-DCLANG_INCLUDE_PATH=${llvmPackages.libclang.dev}/include"
-    (lib.cmakeBool "WITH_CPU_BACKEND" ompSupport)
-    (lib.cmakeBool "WITH_CUDA_BACKEND" cudaSupport)
-    (lib.cmakeBool "WITH_ROCM_BACKEND" rocmSupport)
-    (lib.cmakeBool "WITH_OPENCL_BACKEND" openclSupport)
-  ];
+  cmakeFlags =
+    [
+      "-DCLANG_INCLUDE_PATH=${llvmPackages.libclang.dev}/include"
+      (lib.cmakeBool "WITH_CPU_BACKEND" ompSupport)
+      (lib.cmakeBool "WITH_CUDA_BACKEND" cudaSupport)
+      (lib.cmakeBool "WITH_ROCM_BACKEND" rocmSupport)
+      (lib.cmakeBool "WITH_OPENCL_BACKEND" openclSupport)
+    ]
+    ++ lib.optionals rocmSupport [
+      "-DHIPCC_COMPILER=${finalAttrs.rocmMerged}/bin/hipcc"
+      "-DROCM_PATH=${finalAttrs.rocmMerged}"
+    ];
 
   # this hardening option breaks rocm builds
   hardeningDisable = [ "zerocallusedregs" ];
@@ -92,6 +109,7 @@ stdenv.mkDerivation (finalAttrs: {
   meta = with lib; {
     homepage = "https://github.com/AdaptiveCpp/AdaptiveCpp";
     description = "Multi-backend implementation of SYCL for CPUs and GPUs";
+    mainProgram = "acpp";
     maintainers = with maintainers; [ yboettcher ];
     license = licenses.bsd2;
   };